In the Way Too Early department, I’ve used my Android app to calculate 2015 playoff probabilities based on last year’s final rating and this year’s schedule. Here’s what I came up with using 100,000 simulations of the schedule: AFC East: … Continue reading → (http://www.solecismic.com/frontierblog/?p=78)

More... (http://www.solecismic.com/frontierblog/?p=78)

Overall thinking seems solid. So, let's quickly descend into the minutia.

-Houston over Indy is a curious prediction. I'm fine with someone caling that, but it's hard to see what model assumptions make that the chosen outcome.

-That seems like a lot of love for Dallas. Vegas makes them a half game favorite over Philly in the O/U wins betting... much bigger delat here.

-Overall, the bottom rung teams seem to be too lowly regarded here... 0-3% to make the playoffs? I'd run to put down real money at those implied odds, wouldn't you?

It's completely based on running the 2015 schedule based on point spreads calculated from last season's end ratings.

Houston over Indianapolis - both teams actually ended 2014 with the same rating. Houston's schedule is a tad easier (Kansas City instead of Denver, mainly).

My ratings had Dallas on top of the NFL last season. And Philadelphia far below other teams with similar records. Maybe there's too much reward for road wins and not enough penalty for home losses.

Regarding the bottom teams... Of course these could not be used for Vegas. For three reasons:

1) If odds are high enough, people bet on them, and you have a hard time offsetting those bets. So Vegas always lowers the odds of something with a rare chance of occurring.

2) Elite teams (both good and bad) generally regress a little to the mean the following season. This algorithm assumes that Tennessee is a true 2-14 team. A true 2-14 team has pretty much no chance of putting together 7/8 more wins than it should.

3) You don't know what you're going to get with new players and/or a new season. So you always hedge upward if improvements are made at troubled positions. Just randomly stumbling through 1.6 million games for each team assuming the same point spreads is very different from creating a prediction curve based on historic team performances and randomly stumbling through 100,000 widely variant seasons. That would be a more sophisticated approach, and one I would use if I were creating a model for the type of use a Vegas book would require.

Thanks for the explanation. Actually, just a sprinkling of "regress to the mean" would likely dash my main eyebrow-arching.

In the fine print...so at the end of the season last year, had there been an IND vs HOU game on a neutral field, your system would have called that a tossup? Is that what you mean with your ratings? (I assume so) Have you ever done any look-back on your system vs, available point spreads? Seems like an obvious test to use on any sort of power ratings. I'm curious.